Object detection technology plays a crucial role in computer vision. Although existing YOLO-series models achieve excellent real-time detection performance, they still face challenges in multi-scale feature fusion, channel dependency modeling, and feature reconstruction during up-sampling. To address these issues, this study proposes a multi-scale hybrid feature enhancement object detection network, named MHF-YOLO, which further improves detection performance and robustness by incorporating multi-scale feature fusion and an efficient up-sampling mechanism. Specifically, a multi-scale feature fusion module (MFFM) is designed to leverage a multi-branch dilated convolution structure to efficiently extract and integrate cross-scale contextual information. Furthermore, the mixed local channel attention (MLCA) mechanism is incorporated into the MFFM to optimize feature selection and enhance the network’s response to key regions. Moreover, an efficient up-sampling module (EUSM) is introduced to replace traditional up-sampling methods. By combining depthwise convolution with channel shuffle operations, the proposed module reconstructs spatial details more effectively. Experimental results on the PASCAL VOC 2012 and MS COCO datasets demonstrate that the proposed method outperforms existing YOLO-series models and other object detection models.